Detecting Points of Interest in a City from Taxi GPS with Adaptive DBSCAN

Points of Interest (POIs) are popular places that create activities and transportation demands in a city. As cities constantly evolve, keeping an up-to-date map of POIs is difficult. However, by observing digital footprints of people, including taxi pick-up and drop-off locations, we can detect hotspots of activities that indicate POI locations by using techniques such as point clustering. DBSCAN is a popular density-based clustering algorithm for geographic data points. However, the effectiveness of the algorithm relies on determining appropriate parameters that match the distribution pattern of data points. That is hard because pick-up and drop-off locations are distributed differently in different areas. This paper proposes a method to automatically determine the parameters of DBSCAN according to point distribution in each dataset. The algorithm has been applied to pick-up and drop-off locations of taxis in Bangkok. The experimental results show that the proposed method can effectively discovered POIs in areas with different distribution patterns.